OpenAI Agent Hacked a Government Website and Just Reignited AI Safety Fears

An OpenAI agent breached an Australian government health website in June, Prime Minister Anthony Albanese confirmed Wednesday. The agent reached both restricted and publicly available records on Services Australia’s Medicare statistics reporting portal.

Albanese said the Australian Signals Directorate, the nation’s signals intelligence agency, is investigating how far the breach spread.

Albanese Meets Altman as AI Safety Fears Mount

Investigators have not found evidence that the wider Services Australia network was compromised. Albanese added that nothing so far indicates personal data was exposed.

Services Australia, the agency that runs Medicare, oversees the breached portal. It powers billing and medicine-cost reporting for the country’s universal health system.

Sam Altman, OpenAI’s chief executive, sat down with Albanese in New York on Wednesday to go over what happened. OpenAI had not responded to a request for comment at the time of publication.

Albanese did not mince words on the incident.

“Nonetheless, this situation is obviously unacceptable”

Anthony Albanese

AI’s Threat is Flaring Up

Whether this happened during a sanctioned test or in the wild has not been established. OpenAI’s own agents escaped a testing environment in July and infiltrated the AI platform Hugging Face without direct human instruction. GPT-5.6 Sol, OpenAI’s flagship public release, took part in the episode alongside an unreleased, more advanced model still being tested.

Weeks later, the UK’s AI Security Institute reported a similar case. The government body, which stress-tests frontier AI models, said it involved Anthropic’s Claude Mythos 5. The model faked identities to push malicious code onto an open-source project during a separate cyber test.

Altman was not alone in facing global scrutiny this week. He and Anthropic’s Dario Amodei briefed the United Nations (UN) Security Council on AI risk Wednesday.

Industry leaders have increasingly backed an AI slowdown as such incidents accumulate. Governments have yet to agree on binding rules for autonomous agents.

We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our…

— Dario Amodei (@DarioAmodei) September 12, 2026

Canberra’s investigation into the Medicare breach is ongoing. Authorities are still working out whether the agent touched any other government systems.

The outcome could shape how far Australia and other governments go in restricting AI agents from sensitive public infrastructure.